Abstract
Artificial intelligence (AI) is increasingly transforming intensive care medicine by enabling advanced analysis of complex clinical data generated in intensive care units (ICUs). This review explores current and emerging applications of AI in ICU practice, including sepsis prediction, mechanical ventilation management, acute kidney injury (AKI) forecasting, haemodynamic monitoring, and prognostication. AI-based models have demonstrated the ability to improve early detection of complications, support clinical decision-making, and optimise resource utilisation. However, challenges such as limited interpretability, data integration constraints, and the need for prospective validation continue to hinder widespread clinical adoption. A comprehensive narrative review was conducted using publications from January 2015 to June 2025. Combinations of the terms "artificial intelligence", "machine learning", "deep learning", "intensive care unit", "critical care", "clinical decision support", and "sepsis prediction" were used to search PubMed, Scopus, and Google Scholar. Peer-reviewed original research, systematic reviews, and meta-analyses reporting on the practical uses or clinical validation of AI tools in ICUs were given precedence, while studies focusing solely on algorithm development without clinical integration were excluded. Sepsis, mechanical ventilation, AKI, haemodynamic monitoring, and prognostication are among the thematic areas of application that organise the review. AI has shown significant utility across ICU domains, including early prediction of complications, forecasting mechanical ventilation duration, risk stratification, haemodynamic instability alerts, and mortality prognostication. Models trained on real-world ICU datasets have demonstrated high predictive accuracy and potential for early intervention. However, challenges such as model interpretability, data fragmentation, and ethical concerns remain.